Knowledge-graph view is unreadable when the matrix is large — no node cap, no clustering

What happens

The world map renders every endpoint in the coverage matrix as a node, with no cap, no clustering and no filtering. With the matrix above that is ~620 nodes (561 endpoints + 59 params), of which 360 are paths that do not exist.

The result is a solid field of identical blue nodes labelled GET /.git/config, GET /.env.sql, GET /.htaccess.old, GET /appeal.conf ... The genuinely interesting nodes — the target host, Nginx, HSTS, authenticity_token, client_id, group_id, query — are visually indistinguishable from the noise. (Screenshot available; the operator's words were "that thing became unreadable".)

Where

core/graph/build.py:63 _add_endpoint_nodes iterates matrix["endpoints"] wholesale. The only limit in the graph package is core/graph/views.py:71,94 (next_targets, unrelated). There is no cap on rendered nodes.

Why this matters

The graph's entire value is visual — reachability, data flow and attack chains that the flat matrix cannot express (per core/graph/__init__.py:5). A view that degrades to mush is not a degraded feature, it is a removed one. And it fails exactly when it is most needed: on a large, real engagement.

Expected

The map stays legible regardless of matrix size.

Suggested fix

Independent of the data fix in the companion issue, the renderer should be robust to a noisy matrix:

• cap rendered endpoint nodes (top-N by interest: has findings, has params, has tested cells) with a "+N more" affordance;
• cluster endpoints by normalised path prefix, expandable on click;
• default-hide endpoints with 0 params and 0 tested cells — these are dead weight by definition;
• optionally a provenance filter in the UI, once discovered_by is trustworthy again.

Start at core/graph/build.py:63, where _add_endpoint_nodes iterates matrix["endpoints"], and review core/graph/views.py:71,94 to distinguish existing limits from graph rendering. Read core/graph/init.py:5 for the graph’s intended value. Done means large matrices remain legible through a defined combination of capping, clustering, filtering, and an expansion affordance.